| Abstract: | In a three-choice discrimination task, pigeons learned to distinguish each letter of the alphabet from all the other letters and each of 16 random dot patterns from all the others. Discrimination errors were used to generate a matrix of interletter and interpattern similarities. Human estimates of letter similarity were obtained from the literature, and human subjects rated the similarity of the dot patterns. Pigeon and human performances were described and compared through correlation, multidimensional scaling, and cluster analysis. Fits of the data by simple-feature and template models were computed and compared. The correlation between pigeon and human similarity matrices was .68 for letters and .72 for dot patterns. The other analyses revealed broadly similar patterns of results from the two species but suggested also that, relative to human data, the best fits to the pigeon data required fewer dimensions, fewer features, and fuzzier templates. There was some indication that pigeon discriminations depended on relatively simple features, and several of these were tentatively identified. The different methods employed might have influenced these apparent differences between pigeons and humans, but, overall, the results suggest considerable cross-task and cross species generality in the processing of these simple forms. |
